Abstract: To improve the accuracy of wind farm power forecasting and address the limitations of traditional methods in modeling complex spatiotemporal features and turbine operating condition variations, this paper proposes a dual-model fusion forecasting method based on a grouping strategy. First, the Random Forest algorithm is employed to preprocess the raw data, and a spatial meteorological model is constructed using data from surrounding weather stations and turbine location coordinates. Second, diverse spatiotemporal features are incorporated to enhance the input data representation. The turbines are then clustered according to their operating condition differences, with each group trained separately using LightGBM and XGBoost models, followed by an error-weighted fusion strategy. Experimental results demonstrate that the proposed method achieves significantly higher accuracy compared to conventional single-model approaches, validating its effectiveness and practicality. Through multi-source data integration and a grouped fusion mechanism, this study provides a generalizable framework for high-precision wind power forecasting.